Research map / Label-efficient and robust learning / Recognition and long-tail
Fast Multi-Resolution Transformer Fine-tuning for Extreme Multi-label Text Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace fixed-resolution fine-tuning with recursive multi-resolution fine-tuning (vs X-Transformer) | XR-Transformer · Amazon-3M | P@1 and training time: 51.20 -> 54.20 P@1; 542.0 -> 29.3 hours | Helped |
| add cost-sensitive learning via label aggregation (ablation) | XR-Transformer-1 · Amazon-3M | P@1: 50.2 -> 52.6 | Helped |
| add cost-sensitive learning (ablation) | XR-Transformer-1 · Wiki10-31K | P@1: 86.8 -> 88.0 | Helped |
| add cost-sensitive learning (ablation) | XR-Transformer-1 · Wiki-500K | P@1: 77.6 -> 78.1 | Helped |
| add cost-sensitive learning (ablation) | XR-Transformer-1 · Amazon-670K | P@1: 49.1 -> 49.0 | No clear effect |
| add cost-sensitive learning (ablation) | XR-Linear-1 · Amazon-3M | P@1: 46.8 -> 50.1 | Helped |
| multi-resolution fine-tuned transformer instead of off-the-shelf pre-trained transformer embeddings with TF-IDF (ablation) | XR-Transformer · Wiki-500K | P@1: 70.18 -> 79.40 | Helped |
| multi-resolution fine-tuned transformer instead of word2vec+TF-IDF (ablation) | XR-Transformer · Amazon-3M | P@1: 47.51 -> 54.20 | Helped |
| concatenate TF-IDF features with transformer embeddings (ablation) | fine-tuned transformer embeddings · varied label resolution | P@k | Helped |
| increase leaf cluster size (lower label resolution) for fine-tuning (ablation) | non-recursive fine-tuning of BERT · two-layer HLT | P@k | Hurt |
| XR-Transformer vs LightXML | XR-Transformer · Wiki10-31K | P@1: 89.67 -> 88.69 | Hurt |
Closest papers by meaning
- CascadeXML: Rethinking Transformers for End-to-end Multi-resolution Training in Extreme Multi-label Classification (NeurIPS 2022 · similarity 0.92)
- Large Language Models Meet Extreme Multi-label Classification: Scaling and Multi-modal Framework (AAAI 2026 · similarity 0.90)
- LightXML: Transformer with Dynamic Negative Sampling for High-Performance Extreme Multi-label Text Classification (AAAI 2021 · similarity 0.89)
- ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces (ICML 2025 · similarity 0.86)
- OAK: Enriching Document Representations using Auxiliary Knowledge for Extreme Classification (ICML 2024 · similarity 0.82)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.